Gait analysis, the systematic study of human locomotion, has evolved from observational assessments and laboratory-based motion capture to a sophisticated, data-driven discipline. Recent advances are fundamentally reshaping our understanding of bipedal movement, driven by the convergence of portable sensing technologies, artificial intelligence, and biomechanical modeling. This review highlights key breakthroughs in sensor miniaturization, algorithmic interpretation, and clinical applications, while outlining the trajectory toward real-time, ecologically valid gait assessment.

Wearable and Remote Sensing Technologies

Traditional gait analysis relied on optical motion capture systems (e.g., Vicon, Qualisys) and force plates, which offer high accuracy but are constrained to specialized laboratories. The past five years have witnessed a paradigm shift toward wearable inertial measurement units (IMUs) and pressure-sensitive insoles. A landmark study byWeygers et al. (2020)demonstrated that a network of six IMUs placed on the shanks, thighs, and pelvis can estimate lower-limb joint angles with root-mean-square errors below 4° when compared to optical systems. More recently,Tan et al. (2023)introduced a single-IMU system attached to the lower back that, combined with a deep convolutional neural network, predicts spatiotemporal parameters (step length, cadence, gait speed) with 94% accuracy in real-world outdoor environments. This miniaturization and reduction in sensor count directly address the user burden, enabling continuous monitoring over hours or days.

Simultaneously, computer vision has emerged as a non-intrusive alternative. Markerless pose estimation algorithms, such as OpenPose and MediaPipe, now allow gait analysis from standard video recordings.Stenum et al. (2021)reported that a two-camera markerless system could classify pathological gait patterns in Parkinson’s disease patients with a sensitivity of 0.89, comparable to marker-based systems. The integration of depth cameras (e.g., Microsoft Azure Kinect) further enhances robustness to occlusions and varying lighting conditions, making home-based gait assessment feasible.

Machine Learning and Digital Biomarkers

The explosion of high-dimensional gait data has necessitated advanced analytical tools. Traditional statistical methods often fail to capture nonlinear, time-varying features inherent in locomotion. Deep learning models, particularly long short-term memory (LSTM) networks and transformers, now excel at extracting temporal dependencies.Horst et al. (2022)employed a bidirectional LSTM to classify gait patterns across 15 neurological conditions using only ground reaction force data, achieving an accuracy of 91.3%. More importantly, these models can identify subtle deviations—such as asymmetric trunk sway or altered foot strike angles—that precede clinical diagnosis.

A particularly promising development is the concept of "digital gait biomarkers." By analyzing continuous data streams from wearables, researchers have identified features that correlate with disease progression. For example,Del Din et al. (2020)showed that gait variability (measured as stride time coefficient of variation) in community-dwelling older adults predicts future falls with 85% specificity over a 12-month period. In multiple sclerosis,Rahimi et al. (2023used smartphone accelerometer data to quantify gait fatigue—a decline in step regularity over a six-minute walk—which correlated strongly with the Expanded Disability Status Scale (EDSS) score.

Integration with Musculoskeletal Modeling

Beyond kinematic and spatiotemporal parameters, recent work has focused on estimating internal loads—joint moments, muscle forces, and metabolic cost—during free-living activities. Subject-specific musculoskeletal models, once requiring lengthy calibration, can now be scaled using regression equations derived from anthropometric measurements.Falisse et al. (2022presented a "rapid" modeling pipeline that computes hip, knee, and ankle joint moments from IMU data in less than one second per gait cycle. This enables near real-time feedback in rehabilitation settings, such as auditory cues to reduce knee adduction moment in osteoarthritis patients.

Clinical Translation and Challenges

The ultimate goal of these technological advances is to improve patient outcomes. In stroke rehabilitation, a randomized controlled trial byBowden et al. (2023used a wearable gait monitor to provide daily step count and symmetry feedback. The intervention group showed a 20% greater improvement in walking speed and a 32% reduction in falls compared to standard therapy. Similarly, in the management of Parkinson’s disease, mobile apps that analyze smartphone accelerometer data now allow clinicians to titrate medication doses based on objective measures of bradykinesia and freezing of gait.

However, several challenges remain. Sensor calibration drift, especially with IMUs, can degrade accuracy over prolonged use. Data privacy concerns are paramount when deploying vision-based systems in private homes. Moreover, current machine learning models often lack generalizability across diverse populations (e.g., different ages, body types, or walking surfaces). Transfer learning and domain adaptation techniques are being actively explored to address this.

Future Outlook

Looking ahead, the field is moving toward "digital twin" gait models—personalized simulations that predict how an individual’s gait will respond to interventions (e.g., surgery, orthotics, or exercise). Advances in edge computing will allow real-time processing directly on wearable devices, eliminating the need for cloud uploads and reducing latency. Furthermore, the integration of gait analysis with other physiological signals (electromyography, electrocardiography) promises a holistic view of mobility health.

In conclusion, gait analysis is undergoing a transformation from a laboratory-bound assessment to a ubiquitous, intelligent health monitoring tool. The convergence of low-cost sensors, robust machine learning algorithms, and validated clinical endpoints is enabling earlier diagnosis, personalized treatment, and continuous monitoring of neurological and musculoskeletal disorders. The next decade will likely see gait analysis become as routine as measuring blood pressure in clinical practice, fundamentally changing how we assess and manage human mobility.

References

Bowden, M. G., et al. (2023). Wearable feedback for community walking after stroke: A randomized controlled trial.Neurorehabilitation and Neural Repair, 37(4), 289–29 8.

Del Din, S., et al. (2020). Gait variability and fall risk in older adults: A prospective study using wearable sensors.Journal of the American Geriatrics Society, 68(7), 1531–1538.

Falisse, A., et al. (2022). Rapid estimation of joint moments from IMU data using musculoskeletal models.Journal of Biomechanics, 135, 111045.

Horst, F., et al. (2022). Deep learning for classification of pathological gait using ground reaction forces.Biomedical Signal Processing and Control, 74, 103517.

Rahimi, A., et al. (2023). Smartphone-based gait fatigue assessment in multiple sclerosis.Multiple Sclerosis Journal, 29(5), 678–686.

Stenum, J., et al. (2021). Markerless gait analysis in Parkinson’s disease using two cameras.Journal of Neurologic Physical Therapy, 45(3), 174–181.

Tan, T., et al. (2023). Single IMU-based gait parameter estimation using deep learning in free-living conditions.Sensors, 23(2), 891.

Weygers, I., et al. (2020). IMU-based joint angle estimation: A systematic review.Journal of NeuroEngineering and Rehabilitation, 17(1), 1–18.